Relative intersection of confidence intervals rule for sharper restoration of soft x-ray images

We present a novel method for restoration of images of nanostructures obtained with a soft-ray microscope that uses a 46.9 nm soft x-ray laser microscope for illumination. To suppress the noise and to preserve the image sharpness, we develop a method based on pixel adaptive zero-order modeling of the observed object. Neighboring areas of each pixel are selected using the relative intersection of confidence intervals rule and used for restoration. Due to the non-uniform distribution of noise in the images, we use robust spatial noise modeling. The method provides sharp restored images-sharper than competitive approaches. The sharpness is measured using local phase coherence in the complex wavelet transform domain and shows visible improvement of the novel method.